To facilitate well-informed decision making in today’s noisy financial markets, Accrete.AI wanted to automate cognitive tasks to help investors make sense of unstructured data.
In addition to using human expert networks and proprietary deep learning techniques, the company incorporated IBM Watson to create bias-free, contextually adaptive investment tools.
Increases accuracyof actionable insights buried in linguistic nuances
Saves countless hoursof analyzing earnings calls, corporate filings and other data for meaning
Speeds time to marketfor innovative AI solutions while reducing development costs
Business challenge story
The force multiplier of crowd psychology
Outperforming financial markets in the digital age requires deeper insights—into not only the numbers driving trends, but also the psychology underlying these trends. The faster investors can unveil behavioral insights hidden within human interactions, the better equipped they will be to take advantage of new types of market inefficiencies resulting from the explosion in unstructured data.
“Alpha,” a parameter in the capital asset pricing model (CAPM), refers to the end goal of beating market indices on a risk-adjusted basis. The digital revolution has made delivering alpha more difficult than ever because it has amplified the behavioral aspects of markets. “The nature of market volatility has changed, becoming much more idiosyncratic,” says Prashant Bhuyan, Founder and Chief Executive Officer (CEO) of Accrete.AI. “For example, errant tweets can suddenly send a company’s stock price reeling for no rhyme or reason.”
Traditional investment strategies focus on analyzing structured financial data that fit neatly into rows and columns of spreadsheets and relational databases. While these methodologies were adequate in modeling markets prior to the digital revolution, identifying patterns in today’s extremely noisy, emotionally charged markets requires processing and contextualizing vast volumes of unstructured data. “Information overload is the enemy of alpha because it causes biased decision making,” says Bhuyan. “The problem is only worsening because the growth of unstructured data continues to outpace the biological brain’s processing capacity at an exponential rate.”
To generate alpha in the digital age, investors must evolve by adopting artificially intelligent technologies designed to improve the efficiency and accuracy of research and decision-making processes. Today’s fund managers need tools that can augment their ability to detect changes in topical nuances across quarterly earnings calls, find actionable market-moving rumors within disparate digital data sources, identify supply chain risk and opportunity, and parse confusing messaging from the US Federal Reserve (“Fed speak”) in real time.
“Finding alpha is all about independent thinking,” says Bhuyan. “Investors need tools that surface hidden insights that they can use to generate unique hypotheses. At Accrete.AI, our goal is to build tools that automate specific thinking tasks so that our customers can spend their cognitive resources on developing differentiated perspectives.”
After a decade spent running a high-frequency trading firm, in addition to an analytics company that helped fund managers extract alpha from their own data, Bhuyan realized that his systems were increasingly missing critical insights. He also saw that his clients were feeling the pain of information overload. In response, Bhuyan and his team worked for several years to automate thinking tasks so that they could better understand the new idiosyncratic nature of market volatility. He realized early on that in order to train machines to understand natural language in real time, he would have to solve for the problems of context and bias.
After several years of experimentation, Bhuyan and his team came up with an approach called Machine Augmented Collective Intelligence (MACI). MACI is a framework based on an iterative collaboration between domain-specific expert networks and deep learning systems. This differentiated approach enables the extraction of actionable insights buried in the nuances of language with greater accuracy and efficiency than previous human- or machine-driven efforts. Seeing an opportunity to fill a gap in the marketplace, Bhuyan founded Accrete.AI to build smart, MACI-powered investment tools designed to help fund managers apply cognitive technology and address real-world problems stemming from information overload to generate alpha.
Alpha in the age of super intelligence
For each product it develops, Accrete.AI begins by formally defining a pain point that fund managers are experiencing. The company then sources, vets and formally engages domain experts that work independently, using IBM Watson Knowledge Studio to create a bias-free, contextual kernel specific to the problem at hand. This kernel serves as a basis of understanding from which Accrete.AI’s proprietary neural networks learn to apply reason to the problem within the context of that domain. Knowledge Studio kernels are also used to teach IBM Watson Natural Language Understanding (NLU) linquistic nuances specific to inflation, employment and other financially relevant topics.
MACI ingests raw public information from countless digital data sources in real time using proprietary algorithms and IBM Watson Discovery News, which is a pre-enriched dataset updated daily with approximately 300,000 new articles. Applying the bias-free contextual kernel provided by Knowledge Studio for any given Accrete.AI product, MACI filters the raw data according to domain-specific relevance, such as Federal Reserve insights. Seeded with bias-free contextual awareness, the outputs then feed into a proprietary deep-learning engine that adapts to continuously changing conditions.
Using Knowledge Studio, NLU and Discovery News, Accrete.AI integrates proprietary deep-learning technologies and expert networks to address critical issues that fund managers face in financial markets. Accrete.AI’s marketplace of investment tools helps customers focus on making better decisions and stop wasting time trying to process and contextualize endless digital streams of confusing, constantly changing information.
“Leveraging Watson technologies, MACI is the foundational framework underpinning our offering portfolio. Together, our smart investment tools work like a superhuman brain that’s continuously reading raw information and accreting knowledge. Our vision is that the insights of individual products in our portfolio will eventually compound like interest and bring to light higher order patterns buried in human interaction that would be impossible for any human to discover,” says Bhuyan.
Currently, Accrete.AI’s portfolio features four commercially available products:
- The Accrete.AI Rational Exuberance product parses through Federal Reserve communications, key economic releases and other relevant sources for insights needed to better forecast interest rate changes.
- The Accrete.AI Rumor Hound product continuously reads thousands of data sources to help identify merger and acquisition rumors in real time, validating them according to credibility.
- The Accrete.AI Topic Deltas product monitors changes in nuance and sentiment in key topics referenced across public companies’ earnings calls and press releases.
- The Accrete.AI Data Distillery product analyzes company filings, websites, annual reports and other sources for insights into how weather and other geographical events could affect supply chain and production operations.
Accrete.AI built its smart investment tools on IBM Cloud™ computational infrastructure. The company applies a custom domain ontology, or annotation model, using Knowledge Studio on millions of domain-specific, non-standardized digital data sources, gaining a powerful insight engine tuned to the needs of investors. Using proprietary algorithms, Accrete.AI analyzes millions of articles, chat rooms, industry blogs, social media platforms and other unstructured data from hundreds of thousands of sources daily.
Accrete.AI’s innovative approach to AI includes both supervised and unsupervised learning techniques. The independent domain experts engaged by the fintech startup provide supervised learning, creating bias-free, domain-specific vocabularies, or ontologies, around particular issues. “Experts tag important concepts, entities and relationships in training data to form bias-free ground truths,” says Bhuyan. “These ground truths teach Watson how to interpret the language of a particular domain.”
Unsupervised learning occurs when the outputs from training Watson offerings—such as Knowledge Studio—are fed into Accrete.AI’s shallow- and deep-learning neural networks. These networks identify linguistic nuances related to, but not explicitly defined by, human experts.
Combining these learning techniques, Accrete.AI delivers smart investment tools designed to continuously read and assess official Fed communications and relevant market chatter, find actionable mergers and acquisitions (M&A) rumors buried under mountains of digital data, identify deviations in topical sentiment across quarterly earnings calls, and spot supply chain risk and opportunity before others.
Generating deeper insights
With Watson and IBM Cloud services, Accrete.AI developed and launched sophisticated cognitive investment products designed to help fund managers overcome biases caused by information overload, make better decisions and generate alpha in the digital era. The company’s growing portfolio delivers actionable insight with accuracy and efficiency, surpassing what either humans or machines could achieve independently.
“One of Accrete AI’s fundamental concepts is the ‘glass box,’” Bhuyan explains. “To trust the outputs of an artificially intelligent system, users need to understand the reasoning behind the outputs. For example, in addition to delivering scores that represent an M&A rumor’s credibility, we also provide the snippet that we classified as an M&A rumor and its source. We want customers to use their own intelligence and experience to validate the accuracy of our outputs. Instead of taking our word for it, we would like users to see for themselves how we are able to deliver insight well beyond what a team of analysts could extract.”
“It’s impossible for even an army of skilled human analysts to continuously read every piece of information related to a particular problem necessary to make the most informed decision,” Bhuyan continues. “Our core value proposition is that our tools improve the efficiency of our customers’ investment research processes. Our secondary value proposition is that as our products extract actionable insights buried in unstructured data, we model the data, uncover predictive patterns and share them with our customers.”
The business also benefits from its Embedded Solution Agreement (ESA) with IBM. As a cloud services leader with global reach, IBM provides Accrete.AI with access to high-performance infrastructure, advanced technologies and distribution for its APIs through the IBM Cloud for Financial Services platform. In addition, Accrete.AI is a member of the With Watson program, which gives exclusive brand, marketing and technical resources to developers and organizations that embed Watson technologies in their applications.
Using the IBM Cloud toolset, Accrete.AI can continue to speed time to market with new offerings while reducing development costs. “Our partnership with IBM not only gives us powerful infrastructure to be able to do what we do, but also gives IBM a domain-learning layer and entrée into a new market,” says Bhuyan. “Only a handful of cloud providers have the resources to crunch as much data as we are, and IBM is one of them. As we look to the future in which we are crunching not just text-based natural language but also images, audio and video, we believe that our partnership with IBM is a true competitive advantage.”
Based in New York City, New York, Accrete.AI offers smart, bias-free, contextually adaptive investment tools for a wide range of financial services providers, including hedge funds, trading firms, broker dealers and registered investment advisors. The company’s products help analysts and portfolio managers extract actionable insights buried in the nuances of communication with human-level accuracy. Founded in 2017, the startup employs a global team of 30, including world-class technologists, AI scientists, developers and industry veterans.
Take the next step
To learn more about the IBM solutions featured in this story, please contact your IBM representative or IBM Business Partner, or visit the following websites:
- IBM® Cloud™
- IBM Watson® Discovery News
- IBM Watson Knowledge Studio
- IBM Watson Natural Language Understanding
- With Watson™
The With Watson Program is a global client success initiative that can help accelerate your AI vision through exclusive access to Watson experts and education. With Watson, learn how to enhance your products and applications while growing the power of your brand. To find out more, visit us at: www.ibm.com/watson/with-watson/